How to 3D Scan a Room: LiDAR Apps, Photos or Video

Before you 3D scan a room, decide what you need

People who want to 3D scan a room usually want one of two things, and they are not the same. Some want numbers: wall lengths, a floor plan, something to hand to a contractor or load into a design program. Others want a picture they can move through: the room as it really looks, with its light, clutter and textures, to keep or to show someone.

No single method does both well. There are three practical ways to scan a room today:

• A lidar scanner app on a phone or tablet, for floor plans and measurements.
• Photogrammetry, for a textured mesh you can edit.
• Video to a Gaussian splat, for a realistic walk-around view.

This guide covers how to 3D scan a room with each one, what file comes out, and how to walk the room so the capture works.

Method 1: a lidar scanner app for floor plans and measurements

On Apple devices, most room scanning starts with RoomPlan. It is a Swift API from Apple, introduced with iOS 16, that app developers build into their own apps. Apple describes it as using the camera and LiDAR Scanner on iPhone and iPad to create a 3D floor plan of a room, including dimensions and types of furniture.

According to Apple's documentation, RoomPlan identifies walls, windows, openings and doors, and recognizes furniture and appliances such as a fireplace, a bed or a refrigerator. The result is parametric data: separate components with sizes, not one lump of triangles. It can be exported as USD or USDZ and adjusted in USDZ-compatible tools; Apple names Cinema 4D, Shapr3D and AutoCAD.

Apple also states the limits. In its developer session on RoomPlan, it says the API:

• Works best for a single residential room of up to about 30 by 30 feet, or 9 by 9 meters.
• Needs light, with 50 lux or more recommended.
• Is supported on LiDAR-enabled iPhone and iPad Pro models.
• Is best kept to scans under 5 minutes, to avoid draining the battery and heating the device.

Among third-party apps, Polycam offers a 3D room scanner that uses LiDAR. Polycam says furniture and appliances are detected automatically, that a floor plan with measurements can be generated from the scan, and that standard interior captures are accurate to within a 0.5 inch tolerance. Export formats and floor plan features depend on the plan you pay for, so check before a job depends on it.

The output of this route is tidy and measurable. It is also simplified. A RoomPlan-style model shows a box where your sofa is. It does not show your sofa.

Method 2: photogrammetry for a textured mesh

Photogrammetry rebuilds geometry from overlapping photos and gives you a textured mesh. That is useful if you want to edit the room in Blender or bring it into a game engine. Verified options include RealityScan from Epic Games, which creates 3D models from images and laser scans and is free for individuals and small businesses under US$1 million in annual gross revenue, and Meshroom, an open-source project under the MPLv2 license whose prebuilt downloads target Windows and Linux with CUDA.

Rooms are harder for photogrammetry than objects. The method depends on finding the same visual detail in several photos, and interiors are full of surfaces with none. The documentation for COLMAP, a widely used structure-from-motion tool, says to avoid completely texture-less images and gives a white wall and an empty desk as examples. It suggests adding posters or other background objects if the scene lacks texture. Painted walls and ceilings often come out lumpy or missing.

Photo-only reconstructions also have no built-in real-world scale. Apple's photogrammetry documentation notes that real size can be calculated when images carry depth data, and that otherwise you may have to scale the result yourself. If you need dimensions, measure one wall with a tape and scale the model to match.

Method 3: video to a Gaussian splat for a realistic view

Gaussian splatting takes the same kind of input, photos or a video, but produces something different. Instead of a mesh, the scene is represented by a large number of 3D Gaussians, soft colored blobs, starting from the sparse points found while working out the camera positions. The method comes from a 2023 SIGGRAPH paper by Kerbl, Kopanas, Leimkühler and Drettakis.

For a room, the practical result is a view you can orbit and move through that looks like the footage it came from, including the soft and messy things a floor plan leaves out: fabric, plants, books, the way light falls across a floor.

What it does not give you is a measurable model. There are no walls as objects, no dimensions and no floor plan. Treat it as a realistic record of the room, not a survey.

Several tools make splats. Scaniverse from Niantic Spatial generates Gaussian splats and meshes on the phone itself. On a computer, desktop apps such as SplatLoom work from photos or a video you have already recorded.

Prepare the room first

Ten minutes of preparation saves a rescan.

• Turn on the lights. Apple recommends at least 50 lux for RoomPlan, which it describes as typical for a family living room at night. Brighter is better for photo and video methods too.
• Open the curtains if you are making a lidar scan. Apple notes this lets in more natural light and reduces window occlusions.
• For photo and video methods, keep the lighting the same from start to finish. COLMAP's guidelines ask for similar illumination across images and warn against high dynamic range scenes, such as shooting toward the sun or through doors and windows.
• Stop anything that moves. Send pets and people out, switch off the ceiling fan and the television.
• Clear the floor enough that you can walk the whole perimeter without looking down.

The capture path: how to walk the room

The single most common mistake is standing in the middle and turning in a circle. It feels natural and it gives the software almost nothing to work with. COLMAP's documentation is direct about this: do not take images from the same location by only rotating the camera, and make a few steps after each shot. Depth is worked out from the difference between viewpoints, so the camera has to travel.

A path that works for photos or video:

• Walk the perimeter with your back near the wall, pointing the camera across the room at the opposite side.
• Move slowly and steadily. Blurred frames are wasted frames.
• Make a second lap at a different height, for example chest level and then above your head tilted slightly down.
• Step into each corner and let the camera sweep across it while you keep moving, so both walls and the join between them appear in several frames.
• Overlap generously. COLMAP's guideline is that everything should be seen in at least three images, and more is better.
• Finish where you started, so the first and last frames show the same part of the room.

With a lidar scanner app the rule is simpler: follow the on-screen guidance. RoomPlan's built-in interface shows scanning progress over the real walls and tells you when it needs a different movement or angle.

Mirrors, windows and blank walls

Every method has trouble with the same few surfaces.

Mirrors and glass. Apple says full-height mirrors and glass pose a challenge for the LiDAR sensor. Photo-based methods struggle as well: COLMAP's guidelines say to avoid specularities on shiny surfaces. A mirror shows a second room that is not there, and a splat or mesh may try to build it. Cover large mirrors if you can, or accept that the area will need tidying afterwards.

Windows. Daylight through a window is far brighter than the room, which is exactly the high dynamic range situation COLMAP warns about. Scan when the outside light is soft, or expose for the room and let the window blow out.

Blank and dark surfaces. Plain white walls give photo methods nothing to match. Very dark surfaces can be hard for LiDAR to scan, and high ceilings can exceed the sensor's range, both according to Apple.

None of this makes a room unscannable. It tells you where to look first when a result has holes or ghosting.

Which method for which job

Match the tool to the output you need.

• Measurements, a floor plan, furniture layout, a file for CAD or interior design software: a lidar scanner app built on RoomPlan or a similar system. Output is a simplified, dimensioned model.
• An editable textured model for Blender or a game engine: photogrammetry. Output is a mesh, and you set the scale yourself.
• A realistic view of how the room looks, for a record before renovation, a listing, a memory of a place, or a set reference: video to a Gaussian splat. Output is a splat scene, with no measurements.

These combine well. A lidar floor plan for the numbers and a splat for the look takes two passes around the same room.

Where SplatLoom fits

SplatLoom covers the third route only. It is a desktop app for Mac and Windows that turns overlapping photos or a phone video into a Gaussian splatting scene on your own computer, with no upload and no account. You can crop away stray parts, then export a Gaussian PLY, a PNG still, an MP4 orbit video or a self-contained offline HTML viewer to send to someone.

It does not measure, draw floor plans or export meshes. If you need dimensions, use a lidar scanner app. Results also depend on the capture: mirrors, glass and moving subjects are hard for it, as they are for the method in general.

If what you want is the room as it looks, kept on your own machine, it is a one-time US$19.90 purchase with no subscription.